Opened 14 years ago

Closed 11 years ago

#2696 closed Bugs (fixed)

max_size for boost pool

Reported by: Nikolay Mladenov Owned by: Chris Newbold
Milestone: Boost 1.38.0 Component: pool
Version: Boost 1.37.0 Severity: Problem
Keywords: Cc: boostpool@…

Description

The malloc_needs_resize, ordered_malloc_needs_resize and ordered_malloc members of the pool class double the size of the chunk requested from the operating system every time they cannot satisfy request.

Consider a scenario in which the client allocates about 900Mb before releases the initial size of the pool is 128Mb. on the 1st request from the client the pool will request 128 MB from the OS when the client uses 128M -> 256MB will be requested from the os when the client uses 384M -> 512MB will be requested from the os when the client uses 896M -> 1GB will be requested from the os (which is already likely to fail)

when the client reaches 900M the pool will have close to 2Gb memory allocated from the OS.

But the client will be perfectly happy if the pool just allocates 128M from the OS when it is exhausted, provided 128Mb is large enough to satisfy any single request from the client.

The supplied patch allow for a cap on the chunks the pool is allocating from the OS. This is done by adding a default last parameter max_size which when left at the default of 0 leads to the uncapped behavior.

Attachments (2)

pool.patch (9.3 KB ) - added by Nikolay Mladenov 14 years ago.
boost-pool-maxsize-patch-issue2696.diff (11.9 KB ) - added by Dean Michael Berris 12 years ago.
Update to original patch to apply to r66834 -- tested locally without failures, GCC 4.4 Linux.

Download all attachments as: .zip

Change History (10)

by Nikolay Mladenov, 14 years ago

Attachment: pool.patch added

comment:1 by Steven Watanabe, 14 years ago

Component: Nonepool
Owner: set to Chris Newbold

by Dean Michael Berris, 12 years ago

Update to original patch to apply to r66834 -- tested locally without failures, GCC 4.4 Linux.

comment:2 by John Maddock, 12 years ago

Cc: john@… added

Note to all:

Marshall Clow plans to apply this to Trunk (and then release) to make sure it gets into 1.46. We still need a test case for this though...

comment:3 by John Maddock, 12 years ago

Cc: boostpool@… added; john@… removed

comment:4 by Marshall Clow, 12 years ago

(In [67752]) Applied patch; refs #2696

comment:5 by Marshall Clow, 12 years ago

Resolution: fixed
Status: newclosed

(In [67958]) Merging fixes to release; Fixes #2696; Refs #4346; Refs #2941; Refs #386

comment:6 by John Maddock, 12 years ago

Resolution: fixed
Status: closedreopened

We should update the docs and tests for this issue - reopened for now so we don't forget :)

comment:7 by John Maddock, 12 years ago

(In [69236]) Some further more advanced fixes for issue #2696: Applies existing fix to array allocation, Adds a test case, Adds a fallback mechanism, so that if allocation fails we shrink the next chunk size. Refs #2696.

comment:8 by John Maddock, 11 years ago

Resolution: fixed
Status: reopenedclosed

(In [73495]) Merge updated Pool lib from trunk. Fixes #1252. Fixes #2696. Fixes #4960. Fixes #5526. Fixes #5568. Fixes #5700.

Note: See TracTickets for help on using tickets.